Authors: (including presenting author): :
Chong YM(1), Chan LHD(1), HT Lau (1), Chin LHQ(1), Cheung CWS(1)
Affiliation: :
(1) Department of Radiology, Queen Mary Hospital
Keyword 1: :
Paediatric Musculoskeletal Radiology
Keyword 2: :
Large Language Model
Keyword 4: :
Atlantoaxial rotary fixation
Keyword 5: :
Artificial intelligence
Introduction: :
The rapid evolution of artificial intelligence (AI) and large language models (LLMs) has created new opportunities and lowered barriers for clinicians to develop practical tools that enhance clinical decision-making. This translational project documents our experience exploring how LLMs, coupled with vibe coding, can empower frontline staff to address unstandardized, niche diagnostic challenges. We used the development of a web-based diagnostic calculator for atlantoaxial rotary fixation (AARF)—a rare cervical spine condition that presents interpretive challenges to many radiologists—as a prototype to demonstrate the principles and technical considerations in utilisation of AI-based technology in designing and deploying products addressing clinical challenges. This experience highlights critical operational priorities: precise formulation of clinical questions, data security governance, and thoughtful product delivery architecture.
Objectives: :
This initiative aimed to demonstrate how frontline clinicians can systematically develop AI-enabled tools to enhance efficiency and diagnostic support in subspecialty areas using LLM technology, regardless of prior technical expertise. The specific objectives included: Identifying clinical needs, Evaluating delivery architectures, Designing and developing the product using LLM and vibe coding methodologies, with iterative refinement. Establishing troubleshooting and maintenance protocols to ensure sustained operational reliability and continuous improvement. The presentation illustrates these core components through the development of an AARF diagnostic calculator, which assists frontline radiologists in structured image interpretation and systematic classification of this rare cervical spine pathology.
Methodology: :
The project followed a three-stage systematic process:
1. Preparation and needs assessment. Identification of clinical gaps through direct communication with frontline staff and literature review, resulting in a pragmatic formulation of the clinical question and desired tool specifications.
2. Model design and development. Definition of core product requirements, including system architecture, data privacy safeguards, clinical data fidelity standards, and integration pathways with existing institutional systems. Development proceeded in phases, from proof-of-concept, minimum viable product to final product.
3. Evaluation and iterative refinement. Validation using existing cases and reports, user engagement through interactive sessions, and systematic collection of feedback from frontline staff to inform product improvements. This approach acknowledges the evolving nature of both clinical science and operational requirements.
Result & Outcome: :
The AARF diagnostic calculator successfully demonstrated the value of combining LLM reasoning with vibe coding for frontline clinical support. The calculator provided structured interpretive guidance for this rare cervical spine pathology. Junior radiologists using the web-based version reported both increased diagnostic confidence and improved conceptual understanding of AARF—a condition inconsistently and ambiguously described in literature. The step-wise reasoning interface fostered reflective learning by transparently explaining the diagnostic rationale, serving both educational role in radiology training and clinical role in patient management. The web-based deployment architecture successfully balanced accessibility, data security compliance, and operational maintainability, making it suitable for implementation. This case demonstrates how LLM can be utilised within healthcare operations to address significant but underserved clinical niches—bridging the gap between frontline diagnostic needs and available technological solutions while maintaining quality standards.
